AI agents on your intranet: this is how you arrange findability and management

An HR agent for questions about leave, a sales agent for quotations and an agent who helps to find internal information: more and more applications are emerging within your organization. But do employees also know which agent they can use for what? And who checks whether the answers are still correct?

To use agents properly, you need an overview and clear agreements. Give agents a recognizable place where they can find approved agents, define who is responsible, and align management with what an agent is allowed to do. In this blog you can read how to go about it.

A good agent must also be found

Building an agent is a first step. After that, colleagues must know how to find him and understand when he helps them.

That's exactly where things can go wrong. An agent is shared with a few colleagues, the link disappears in a conversation and the rest of the organization hardly knows it exists. Meanwhile, another team is building something similar.

Get an agent for HR questions. If different departments each make their own version, there are several places that need maintenance. Do they all use the current employee handbook? Does HR know these agents exist? And who can an employee turn to if an answer is incorrect?

A recognizable place for approved agents helps employees choose and makes it easier to reuse existing solutions. For each agent, state what it is intended for, who can use it and who the owner is. This way, employees know better what to expect and where to go with questions.

Integrate agents into their daily work

A central collection of agents is only useful if employees know how to reach them. Therefore, give agents a logical place in the digital workspace that colleagues use every day.

An employee looking for information about employee benefits should also be able to find the corresponding HR agent there. A sales colleague wants to be able to quickly open the customer proposal agent and save it for the next moment. A classification by topic and a personal selection of frequently used agents help with this.

The Debble Agent Dashboard is intended as a central entry point for agents within the digital workspace. In the demonstration version, agents can view agents by topic, add favorites to "My Agents," and start a conversation instantly. That personal selection also remains accessible while reading a news item or visiting another page within Debble.

For example, employees can use an agent when he or she helps them with their work.

Make ownership part of every agent

As soon as colleagues start to rely on an agent, it must be clear who monitors the quality. Who checks the sources? Who processes feedback? And who will take over if the maker gets another position or leaves the company?

Therefore, appoint an owner for each agent and agree on when the operation will be reassessed. Such an assessment moment helps prevent an agent from remaining available while the information or the underlying work process has changed in the meantime.

In any case, take these questions into account during that check-up:

• Is the information on which the officer bases his answers correct?
• Do the links still work as intended?
• Does the officer still fit in with the work process?
• Is it clear who is picking up feedback and implementing improvements?

A management dashboard can show which agents don't have an owner or need to be reassessed. The organization must then agree on who will work on this and when. Also record what happens if an agent no longer meets the agreements, for example temporarily decommissioning until the information is updated.

Align management with what an agent is allowed to do

An agent who looks up information asks for different agreements than an agent who changes data or starts a process independently. The sensitivity of the information and the consequences of an error also count.

You can organize management with three zones:

• Green: Agents experiment with agents for personal productivity within clear boundaries.
• Orange: employees with the right knowledge develop applications for their team or department, with agreements on quality and ownership.
• Red: IT or a specialized management partner manages applications whose impact and complexity require additional expertise.

Determine in advance which boundaries belong to each zone and when an application needs additional assessment. This gives employees space that suits their knowledge and what the agent can do.

For example, a team that knows its own process well can develop improvements itself. As soon as an agent also starts changing data or working with sensitive information, it must be reconsidered which agreements and expertise are needed.

Also include costs in those agreements. Determine who tracks the consumption and how you assess what the application delivers. Time savings and quality are just as relevant as the cost of use.

Start small and improve with real users

Choose a recognizable task, build a first version and let a limited group of colleagues work with it. Their feedback helps to discover where the agent is already supporting well and where improvement is needed.

The maker often knows exactly which question will give a good answer. A colleague formulates that question differently, lacks context or expects something that has not yet been foreseen. It is precisely this use that makes it clear what you need to adjust before you make the agent more widely available.

Therefore, make it easy to give feedback and agree on who will assess it. Look at both the answers and the use: can employees find the agent, do they understand what he is intended for and does he help them to perform their task?

Arrange an owner, a suitable target group and a moment to reassess the operation from the start. Then give the agent a fixed place where employees can find him. In this way, development, use and improvement are in line with each other.